QVQ-72B-Preview: benchmark results
Alibaba's experimental 72B visual-reasoning model built on Qwen2-VL-72B, released as a December 2024 research preview. Provider: Alibaba. Released 2024-12-24. Access: Open.
Aggregate Rating 1600 ± 12, rank #440 of 1656 rated models, from 174 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GraphVerse - Maximum Common Subgraph (Paired Images) - VGR-Score | 39.9 | Visual graph reasoning score (0-100; 0.7 process + 0.3 answe | 100 |
| Open LMM Reasoning - MathVerse - Applied | 53.2 | Accuracy (%) | 98 |
| OpenVLM MathVista - STA | 86.7 | Accuracy (%) | 96.6 |
| OpenVLM MathVista - FQA | 77.7 | Accuracy (%) | 96.1 |
| Open LMM Reasoning - MathVerse - Coordinate | 75 | Accuracy (%) | 95.9 |
| GraphVerse - Maximum Independent Set - VGR-Score | 45.3 | Visual graph reasoning score (0-100; 0.7 process + 0.3 answe | 94.4 |
| Open LMM Reasoning - LogicVista - numerical | 84.2 | Accuracy (%) | 93.8 |
| Open LMM Reasoning - MathVision - analytic geometry | 46.4 | Accuracy (%) | 93.4 |
| Open LMM Reasoning - LogicVista | 58.2 | Accuracy (%) | 93.2 |
| Open LMM Reasoning - DynaMath - Subject-analytic geometry; Avg | 56.9 | Accuracy (%) | 92.9 |
| Open LMM Reasoning - MathVerse - Plane Geometry | 51.6 | Accuracy (%) | 92.9 |
| Open LMM Reasoning - DynaMath - Subject-analytic geometry | 22.7 | Accuracy (%) | 92.3 |
Interactive version: theaggregate.ai/model?slug=qvq-72b-preview · How It Works · Data refreshed daily, snapshot 2026-10-20.